Oh - I'm not saying that Lambda is what I'd personally choose, although there definitely seems to be quite a few people that do like it.

As mentioned in the article - the base unit of compute on public cloud is the virtual machine so there are quite a few tools out there to deploy those. Some microvm implementations out there won't boot straight off AWS/GCE but many including Nanos do. It takes ~2min to deploy && boot most applications we are running on GCE - AWS is a bit longer (but I could see that changing in the future). Note - in this case you are actually uploading the entire image - if you are just spinning up/down than it's much much faster.

Yeah, I guess it depends on your goals. Some of the unikernel people talk about VMs even smaller/cheaper than a t3.nano that start faster than any cloud can start a VM. Clearly to run such a thing you would need your own orchestration layer. Maybe that's a niche within a niche though.

Agreed.

I think for our general use case we are looking at people with needs that are using vms or containers today (eg: not necessarily those that want plaid speed boot-times). That's not really something we are currently optimizing for. For something like a database there would be no desire to spin up/down so fast. Also, many frameworks with heavy language vms such as JVM or ruby take a long time to boot regardless so trying to run those in a 5-10ms boot-time doesn't make a ton of sense.

Having said that, there most definitely are very very interesting projects like LightVM, http://cnp.neclab.eu/projects/lightvm/ that do target low latency and NFV would be a great use-case for projects like that.